FSO, FSE and frame pacing in CS2
On this page
Short answer: our measurements found no change in the Raw Input structure or an advantage of FSE in average click-to-photon latency. At the same time, FSO and FSE can affect presentation and frametime distribution differently. Therefore BoosterX recommends testing disabling FSO selectively in a specific shooter, rather than applying it as a universal optimization.
Status: 300 synthetic Raw Input runs on Windows 10 and Windows 11 and a historical physical FSO/FSE series in Valorant are complete. The new click-to-photon CS2 series on the BXLAT map and frametime measurement are ongoing. Third-party video is used only for independent cross-checking.
Claim under test
Section titled “ Claim under test”We are testing two different claims that must not be merged:
- Disabling FSO changes the presentation path and frame time distribution in CS2.
- Disabling FSO changes game Raw Input packets and therefore by itself makes aiming more precise.
The first claim is tested simultaneously by frametime and physical click-to-photon latency. Our completed measurements do not confirm the second claim.
Scope of research
Section titled “Scope of research”Windows 10 22H2 and Windows 11 25H2; CS2 and Valorant; the physical BoosterX click-to-photon rig and synthetic Raw Input runs. Other games, resolutions and builds require separate testing.
What FSO, FSE and presentation modes mean
Section titled “ What FSO, FSE and presentation modes mean”Fullscreen Optimizations (FSO) let Windows keep the appearance of a fullscreen game while using an optimized windowed path. Microsoft describes it as a combination of fullscreen exclusive performance with faster Alt+Tab and overlay support.
Fullscreen Exclusive (FSE) gives the game exclusive control over output. On modern versions of Windows, the windowed flip model can also send frames to the display directly: with Independent Flip, DWM does not have to perform normal composition of every frame.
Legacy Flip and Independent Flip describe how finished frames are delivered to the display. They do not describe the mouse data path. Therefore a change in presentation mode can affect frame pacing and visual feedback without changing Raw Input packets.
Related BoosterX settings
Section titled “Related BoosterX settings”The “Global Fullscreen Optimizations” setting is related to this topic. On Windows 11 25H2, the previous method of applying it did not receive sufficient confirmation, so the setting should be considered experimental and tested only on a specific game.
Game Bar and Game Mode
Section titled “Game Bar and Game Mode”Additional testing separated two independent settings. “Game Bar” controls the system game overlay, Game DVR and the background capture path. “Game Mode” separately enables the built-in Windows mechanism that limits some background activity during a game and prevents certain Windows Update operations.
Disabling Game Bar reliably removes its capture and UI functions, but not third-party overlays, and does not guarantee an FPS increase. Game Mode is a working Windows feature; enabling it is reasonable for a gaming PC, but the final FPS and frame time depend on the game, driver and current bottleneck.
Practical pages: “Game Bar” and “Game Mode”.
- Xbox Support: Use Game Mode, checked 2026-09-01.
- Microsoft: AllowGameDVR policy (Windows Game Recording and Broadcasting), checked 2026-09-20.
What BoosterX has already tested
Section titled “ What BoosterX has already tested”The research is divided into independent layers. This prevents mistaking a change in presentation mode for a change in mouse data or replacing physical latency with a software metric.
| Layer | State | Main result |
|---|---|---|
| Raw Input integrity | Complete | No stable drop, merge or split between modes found. |
| Input → first visible frame in a synthetic environment | Complete | Differences less than 1 ms, direction depends on the scenario. |
| Physical FSO/FSE click-to-photon in Valorant | Complete, historical series | Averages differ by no more than 0.04 ms. |
| Physical FSO/FSE click-to-photon in CS2 | In progress | At least 300 valid clicks per state on the BXLAT map are planned. |
| Frametime CS2 | In progress | Normal frames, percentiles and rare spikes are tested separately. |
Raw Input matrix
Section titled “Raw Input matrix”A synthetic matrix of 300 runs is complete: 60 combinations with five repetitions. Windows 10 22H2 and Windows 11 25H2, windowed, borderless and exclusive modes, native resolution and 1280×960, as well as five mouse movement scenarios were tested.
| Factor | Coverage |
|---|---|
| Windows | 10 22H2 and 11 25H2 |
| Window modes | Windowed, borderless and exclusive |
| Resolution | Native and 1280×960 |
| Input scenarios | Five scenarios, including 1 kHz and micro-jitter |
| Repetitions | Five for each of the 60 combinations |
The first run erroneously showed event loss and coalescing only on Windows 10. After equalizing the allocated processor resources of both test systems, the difference disappeared. Repeated 150 runs of Windows 10 and the original Windows 11 runs showed no drop, merge or split Raw Input events.
This repetition matters for the methodology: the initial difference looked like a Windows-specific trait, but disappeared after the difference in CPU resources was eliminated. Therefore it is excluded from the product conclusion, rather than used as a convenient confirmation of the setting.
Input → first visible frame
Section titled “Input → first visible frame”The measured cross-system shifts of the median from Raw Input event to first visible frame remained less than 1 ms and changed direction between scenarios:
| Scenario | Windows 11 − Windows 10 |
|---|---|
| Constant movement 1 kHz, 32 ms | +0.510 ms |
| Constant movement 1 kHz, 256 ms | +0.472 ms |
| Micro-jitter, 125 Hz | −0.797 ms |
The average difference in the number of frames to a visible result was 0.0 when comparing borderless with exclusive and native resolution with 1280×960. The sign changed between scenarios, so this stage showed no stable advantage of one Windows or one presentation mode.
This part of the research confirms only the behavior of a synthetic application in a controlled virtual environment. It does not measure mouse-to-photon latency, real CS2 frametime or the capabilities of a physical GPU. The stage with real CS2 and external frame capture is not yet complete, so we do not present it as a finished result.
Physical click-to-photon measurements
Section titled “ Physical click-to-photon measurements”BoosterX has also been using a hardware click-to-photon rig for several years. The electrical signal of the left button of a Logitech G PRO X SUPERLIGHT starts an Arduino Uno timer, and a photosensor on the monitor stops it after a pixel changes. The result includes the mouse controller and debounce, USB, Windows, the game, the rendering pipeline, the GPU and the monitor. The detailed scheme and series rules are described in the research methodology.
In the historical Valorant series on Windows 11 24H2, 100 clicks per state were performed. The resolution, computer, monitor and the rest of the configuration within the comparison did not change.
| State | AVG | STDDEV | MIN | MAX |
|---|---|---|---|---|
| FSO, Display scaling | 10.35 ms | 2.25 ms | 5.72 ms | 15.46 ms |
| FSO, GPU scaling | 10.39 ms | 2.43 ms | 5.71 ms | 15.12 ms |
| FSE, Display scaling | 10.39 ms | 2.29 ms | 5.49 ms | 15.68 ms |
| FSE, GPU scaling | 10.39 ms | 2.47 ms | 5.60 ms | 14.34 ms |
In this series, the averages of FSO and FSE differed by no more than 0.04 ms, and the spreads overlapped. It showed no measurable advantage of FSE in click-to-photon. This does not disprove a possible change in frame pacing: the same average click latency can coexist with a different frame time distribution and rare spikes.
The series is historical: it preserves AVG, STDDEV, MIN and MAX, but has no source distribution, P90 or graph. The current protocol requires at least 300 valid clicks per state and extended statistics. A repeated physical series for CS2 on the BXLAT map with FSO/FSE is currently being prepared.
Independent cross-check
Section titled “ Independent cross-check”CS2 Kitchen independently compared FSO and disabled FSO on several computers with Windows 10 and Windows 11, five runs per state. In the configuration shown, disabling FSO moved CS2 from Independent Flip to Hardware: Legacy Flip.
The main result matches the direction of our research: input latency practically did not change, but the frametime distribution became different. Legacy Flip improved most normal frames and P1, while on some systems creating heavier rare spikes. The full source samples were not published, so this is an external corroborating observation, not part of BoosterX statistics.
What this means for the feel of aiming
Section titled “ What this means for the feel of aiming”Raw Input integrity and the feel of control do not contradict each other. The game can receive the same mouse events but show the visual result with a different frametime distribution. A smoother or faster main body of frames can subjectively feel like more predictable aiming.
Our Raw Input and click-to-photon results do not show a direct click speed-up from FSE. At the same time, the change in presentation and frame pacing provides a measurable explanation for player reports of a different feel of aiming. This does not prove an increase in player accuracy: such a conclusion would require separate blind gameplay tests.
What is confirmed
Section titled “ What is confirmed”- FSO changes the way Windows services a fullscreen game; Microsoft allows disabling it for a specific game in case of a regression or input lag.
- The modern flip model can use Independent Flip and output frames directly with efficiency comparable to FSE.
- In our synthetic matrix, presentation mode did not create a stable difference in Raw Input integrity.
- Our historical physical Valorant series showed no advantage of FSE in average click-to-photon latency.
- In the external CS2 test, Legacy Flip improved the time of most frames, but could create rare heavier outliers without a significant change in input latency.
What is not confirmed
Section titled “ What is not confirmed”- That FSE or Legacy Flip is always faster than FSO and Independent Flip.
- That disabling FSO improves aiming accuracy or changes Raw Input packets.
- That the CS2 result automatically applies to Valorant and other shooters.
- That virtual measurements describe physical GPU, display or mouse-to-photon latency.
- That rare microfreezes of Legacy Flip will appear or disappear on a user’s specific computer.
Limitations
Section titled “ Limitations”Our 300-run matrix uses a controlled synthetic load, not a real CS2 match. The historical physical Valorant series consists of 100 clicks per state and contains no source distribution, P90 or graph. The external video uses several computers but does not provide us with a full open set of samples. The GPU configuration, driver, Windows build, overlays, HDR, VRR, FPS limit and background load can change the presentation path and the result.
The research and the tools used belong to the BoosterX developer, who provides the FSO setting, so the developer has a direct interest in the results. The methodology and scope of applicability are described above, and the conclusions can be verified against open data and the listed public sources. The negative Raw Input results and the risk of rare frametime spikes are stated on par with a possible improvement.
Practical BoosterX recommendation
Section titled “ Practical BoosterX recommendation”By default, leave FSO enabled. For a competitive shooter, the setting can be disabled selectively only for a specific game through BoosterX and both states compared. For CS2 there is already an external basis for such a test. Valorant and other games must be measured separately.
Compare the same scene, FPS limit and background load in at least five paired runs. Look not only at average FPS, but also at the frametime graph, P1, 1% low, 0.1% low and individual outliers. The in-game FPS counter is not enough for this.
Leave FSO disabled only if normal frame pacing has improved and new microfreezes, problems with Alt+Tab, overlays, HDR or VRR have not appeared. Use the subjective feel of aiming as an additional observation, not as the only metric.
Restoring the state
Section titled “ Restoring the state”For the global toggle, use the recommendation on the “Global Fullscreen Optimizations” page: globally keep FSO enabled, and test the exception for a specific game.
Return the FSO setting for the game to the default state through BoosterX and completely restart the game. A Windows reboot is usually not required for this comparison. After reverting, make sure the game again uses the original presentation mode.
Public sources
Section titled “ Public sources”- Demystifying Fullscreen Optimizations — the purpose of FSO, the trade-offs of FSE and selective disabling for a game.
- For best performance, use DXGI flip model — DirectFlip, Independent Flip and direct output without normal DWM composition.
- DXGI flip model — the design of the flip model in Windows.
- Public table of BoosterX click-to-photon measurements — historical CS2 and Valorant series.
- Logitech G PRO X SUPERLIGHT review, RTINGS — an independent assessment of the click latency of the mouse used.
- I Found a Weird Way to Make CS2 Run Better, CS2 Kitchen, published 2026-08-21 — an independent cross-check with BoosterX results.
Public sources and wording checked: 2026-08-24.
Change history
Section titled “Change history”- 2026-09-20: the conflict-of-interest disclaimer was strengthened to the full wording with the ownership of the research and tools.
- 2026-08-24: BoosterX research was made the basis of the article; the external CS2 test was moved into a brief independent cross-check.